Weak vibration signal detection based on frequency domain cumulative averaging with DVS system

•For the first time, the use of frequency domain cumulative averaging (FDCA) algorithm is proposed in distributed vibration sensing (DVS) system for the detection of weak vibration signals, and the effects of the length of each segment signal and the number of segments in the FDCA algorithm on the signal-to-noise ratio (SNR) enhancement and variance reduction are also investigated.•Experimentally measured when the SNR of the weak vibration signal is less than or equal to 1, the FDCA algorithm can be used to achieve an improvement of SNR of 6.91 dB. In addition, the FDCA algorithm is used for the detection of weak vibrations generated by self-oscillation and external excitation on a simply supported beam constructed from 3 m-long wooden strips and achieves SNRs of 24.83 dB and 14.1 dB, respectively.•Detection and localization of weak vibration frequencies with high spatial resolution can be achieved through the construction of a three-dimensional map of position-frequency-spectral power.•A good inverse relationship between the amplitude of the applied weak vibration and the length of the sensing fiber is deduced from simulation and experiment, and the result can better guide the selection of the length of the sensing fiber in practical applications to better realize the detection of weak vibration. Accurate detection of weak vibration signals by distributed vibration sensing (DVS) system plays an important role in early warning of large infrastructures. In this paper, a frequency domain cumulative averaging (FDCA) algorithm is proposed to overcome the problem that weak vibrations cannot be identified and localized when the signal-to-noise ratio (SNR) of the weak vibration signals is less than or equal to 1. The proposed method periodically truncates the weak vibration signal in the time domain and performs power spectrum estimation separately. Subsequently, the power spectrum obtained in the frequency domain is cumulatively averaged to improve the SNR of the weak vibration signal and reduce the spectral variance. Finally, the averaged spectrum obtained at each position is used to construct a three-dimensional spectrogram of position-frequency-spectral power, which allows the identification of weak vibration signals with high spatial resolution. The experimental results well demonstrate that the proposed method can detect weak vibration signals acting on a sensing fiber with a length of 1 m, and the SNR is improved by 6.9 dB. In addition, the proposed method